The Effect of Conference Calls on Analysts' Forecasts – German Evidence

Abstract

This study examines whether conference calls provide additional information to analysts. For a large sample of conference calls, hosted by German firms between 2004 and 2007, our results show that conference calls improve analysts' ability to forecast future earnings accurately. This suggests that additional information is released during conference calls. The reduction in forecast error is economically significant and larger in magnitude when compared to results for the US (Bowen et al., 2002). These findings are consistent with the notion that committing to additional disclosures is likely to yield greater effects in a less stringent disclosure system (Verrecchia, 2001). Since the majority of our sample firms conduct conference calls as closed calls, the evidence of this paper suggests that conference calls may contribute to an information gap between call participants and non-invited parties. Our findings should be of substantial interest to European regulators seeking to level the informational playing field for all investors.

Acknowledgements

We thank Salvador Carmona (the editor), Felix Fischer, Günther Gebhardt, Martin Glaum (discussants at the VHB Annual Conference 2010), Monika Hommel, Michael Raab, Michael Scholz, Joao Toniato, two anonymous referees, workshop participants at the EAA Annual Congress 2010, the VHB Annual Conference 2010, and two anonymous reviewers for the VHB Annual Conference 2010 for their valuable comments. Moritz Bassemir is grateful to the Schmalenbach Gesellschaft für Betriebswirtschaft for providing financial support to travel to the EAA Annual Congress 2010. Zoltan Novotny-Farkas gratefully acknowledges the financial contribution of the European Commission Research Training Network INTACCT (Contract MRTN-CT-2006-035850). Part of the data was collected during a research visit to the INTACCT partner institution Lancaster University.

Notes

For instance, in January 2003 Directive 2003/6/EC was adopted. Its objective is to preclude the selective disclosure of information. The directive contains three main elements: insider dealing, ad-hoc publicity and market manipulation. In Germany, Section 15 WpHG (Wertpapierhandelsgesetz, in English: German Security Act) was passed in 1995, requiring listed firms to instantly publish all information that may substantially influence stock prices through ad-hoc press releases. Further, Section 15a WpHG was introduced in the year 2002, establishing that trading activities in a firm's stock by corporate insiders (directors' dealings) must be disclosed.

Following Bushee et al. (2003) and Skinner (2003), we understand ‘open access conferences’ as providing unlimited access to all available information to the universe of capital market participants at the same point in time. For instance, live broadcasts of conference calls via the internet (e.g. webcasts) that are open to all interested parties meet these requirements. Drawing a random sample of 80 firms from our full sample of 352 and calling the firms' investor relations agents in February 2010 revealed that 86% of these firms conduct conference calls as closed calls.

See also Bradshaw (2009) for a schematic of analyst information processing.

In addition, our research design includes several features to ensure that any observable change in forecast properties is primarily attributable to the information released during conference calls and not to differences in the analysts' information processing across conference call quarters and non-conference call quarters (see Section 3).

There is evidence that (1) managers discriminate among analysts by allowing them to ask questions conditional on how favourably the analyst views the firm (Mayew, 2008), and (2) managers withhold information by not answering analysts' questions (Hollander et al., 2010). However, a manager's unwillingness to disclose is also likely to serve as an informative signal to analysts.

Increased public disclosure may also substitute for analysts' private information because public information mitigates the gains from private information (e.g., Kim and Verrecchia, 1991), i.e. conference calls may reduce analysts' incentive to acquire private information (substitutional effect). However, the substitution effect typically occurs when firms make a long-run commitment to greater public disclosure, while the complementary effect is more likely a response to anticipated public disclosure events such as conference calls (Verrecchia, 2001, p. 172).

In the BKLS framework private information is by definition uncorrelated. Since it is unlikely that all the information released by managers in private conversations with analysts is unique, much of this information is most likely considered as public information in the BKLS framework (Bowen et al., 2002, p. 315).

We do not use the I/B/E/S consensus forecast, because we place additional restrictions on the calculation of the forecast error; these are outlined in Table 1 in Section 4.

For our sample (N = 2538 earnings announcements) the average time period between the beginning of the quarter and EA-Q4 _t-1 (i.e. the announcement of the Q4 _t-1 earnings) is 79 days, for EA-Q1 _t : 39 days, EA-Q2 _t : 38 days, and EA-Q3 _t : 39 days.

Bowen et al. (2002) set the POST2 time horizon at the earnings release date of the next interim earnings. Our research design ensures that information coming along with the next interim earnings release is not included in ΔERROR2. Also, a constant time frame of an additional 30 days gives all firms the same amount of time to ‘catch up’.

Results are qualitatively similar if we choose total assets as a proxy for SIZE.

In Germany, conference calls related to the annual earnings announcement are usually held physically, while conference calls in conjunction with quarterly earnings releases are conducted as telephone and/or webcast access conferences. This information was inferred during several informal conversations between the authors and investor relations officers from Prime Standard firms.

However, we control for the age of forecasts in our multivariate regressions. In addition, in sensitivity tests we only include observations for which forecasts are revised within the 20 days following an earnings announcement.

For our sample (N = 2,538 earnings announcements), the average time period between subsequent earnings announcements for EA-Q4 _t-1  → EA-Q1 _t is 50 days, for EA-Q1 _t  → EA-Q2 _t : 91 days, EA-Q2 _t  → EA-Q3 _t : 92 days, and EA-Q3 _t  → EA-Q4 _t : 130 days.

For the combined sample, Bowen et al. (2002), p. 300, report an average initial level of forecast error of 0.4% (of share price). It should be noted that the larger initial level of forecast error for our German sample may also be partly attributable to the use of annual earnings forecasts (which are more difficult to predict due to the longer forecast horizon), while the study of Bowen et al. (2002) for the US setting is based on quarterly earnings forecasts. However, for our German sample, the initial forecast error one-quarter ahead (at EA-Q3 _t ) of the forecasted annual earnings (EA-Q4 _t ) is still 2.6% (of share price).

We obtain this result by dividing the coefficient of CC (0.00187) by the interquartile range of ΔERROR1 (0.00015 – (–0.00345) = 0.00360).

Bowen et al. (2002, p. 304), find that the decrease in forecast error is approximately 0.013% (of price) or 21% of the interquartile range.

SIZE and NO are positively correlated, with a coefficient of 0.76 (not tabulated). We assess the presence of multicollinearity by dropping either SIZE or NO from the regression model. The results remain qualitatively similar. In addition, VIF statistics do not indicate that multicollinearity presents a severe problem in our regressions.

We thank an anonymous referee for bringing this issue to our attention.

If we use alternative cut-off criteria for the ranking of the analysts, such as upper versus lower quartile, our results remain qualitatively similar.

I/B/E/S has stopped publishing the names of brokerage houses and analysts. They only assign individual IDs to each brokerage house and analyst, allowing us to track prior performance.

In these regressions, we include the difference between the low- and high-ability groups for ΔAGE1(2) and ERROR_PRE , similar to our treatment of the dependent variable.

Results are available upon request.

Recall that our sample firms are required by stock market regulations to host at least one conference call per calendar year.

Additional information

Notes on contributors

Moritz Bassemir

Paper accepted by Salvador Carmona.

Zoltan Novotny-Farkas

Paper accepted by Salvador Carmona.

Julian Pachta

Paper accepted by Salvador Carmona.